Ko Trocr

🚀 Korean TrOCR Model

This is a TrOCR model that uses a decoder model with a tokenizer that can handle initial consonants, ensuring that initial consonants are not recognized as UNK. It leverages the know - how gained from the 2023 Teachers' Group AI OCR Challenge.

🚀 Quick Start

✨ Features

• The TrOCR model uses a decoder model with a tokenizer that can handle initial consonants, preventing initial consonants from being recognized as UNK.
• It is developed by leveraging the know - how from the 2023 Teachers' Group AI OCR Challenge.

📦 Installation

No specific installation steps are provided in the original document, so this section is skipped.

💻 Usage Examples

Basic Usage

from transformers import TrOCRProcessor, VisionEncoderDecoderModel, AutoTokenizer
import requests 
import unicodedata
from io import BytesIO
from PIL import Image

processor = TrOCRProcessor.from_pretrained("ddobokki/ko-trocr") 
model = VisionEncoderDecoderModel.from_pretrained("ddobokki/ko-trocr")
tokenizer = AutoTokenizer.from_pretrained("ddobokki/ko-trocr")

url = "https://raw.githubusercontent.com/ddobokki/ocr_img_example/master/g.jpg"
response = requests.get(url)
img = Image.open(BytesIO(response.content))

pixel_values = processor(img, return_tensors="pt").pixel_values 
generated_ids = model.generate(pixel_values, max_length=64)
generated_text = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
generated_text = unicodedata.normalize("NFC", generated_text)
print(generated_text)

📚 Documentation

Train Datasets

• AI Hub
  • Diverse Forms of Korean Character OCR
  • Public Administrative Document OCR

Model Structure

📄 License

This project is licensed under the Apache - 2.0 license.